A biologically inspired approach for interactive learning of categories

An amazing capability of the human visual system is the ability to learn a large repertoire of visual categories. We propose an architecture for learning visual categories in an interactive and life-long fashion based on complex-shaped objects, which typically belong to several different categories. The fundamental problem of life-long learning with artificial neural networks is the so-called “stability-plasticity dilemma”. This dilemma refers to the incremental incorporation of newly acquired knowledge, while also the earlier learned information should be preserved. To achieve this learning ability we propose biologically inspired modifications to the established learning vector quantization (LVQ) approach and combine it with a category-specific forward feature selection to decouple co-occurring categories. Both parts are optimized together to ensure a compact and efficient category representation, which is necessary for fast and interactive learning.

[1]  Fred Henrik Hamker,et al.  Life-long learning Cell Structures--continuously learning without catastrophic interference , 2001, Neural Networks.

[2]  M. Hasselmo,et al.  High acetylcholine levels set circuit dynamics for attention and encoding and low acetylcholine levels set dynamics for consolidation. , 2004, Progress in brain research.

[3]  Shen Furao,et al.  An incremental network for on-line unsupervised classification and topology learning , 2006, Neural Networks.

[4]  W. Scoville,et al.  LOSS OF RECENT MEMORY AFTER BILATERAL HIPPOCAMPAL LESIONS , 1957, Journal of neurology, neurosurgery, and psychiatry.

[5]  Teuvo Kohonen,et al.  Self-Organization and Associative Memory , 1988 .

[6]  Thomas Martinetz,et al.  SoftDoubleMaxMinOver: Perceptron-Like Training of Support Vector Machines , 2009, IEEE Transactions on Neural Networks.

[7]  Heiko Wersing,et al.  A Comparison of Features in Parts-Based Object Recognition Hierarchies , 2007, ICANN.

[8]  M. Hasselmo,et al.  Enhanced cholinergic suppression of previously strengthened synapses enables the formation of self-organized representations in olfactory cortex , 2003, Neurobiology of Learning and Memory.

[9]  Michael J. Swain,et al.  Color indexing , 1991, International Journal of Computer Vision.

[10]  Thomas Villmann,et al.  Generalized relevance learning vector quantization , 2002, Neural Networks.

[11]  I. Izquierdo,et al.  Separate mechanisms for short- and long-term memory , 1999, Behavioural Brain Research.

[12]  Horst-Michael Groß,et al.  A vision architecture for unconstrained and incremental learning of multiple categories , 2009, Memetic Comput..

[13]  R. French.  Catastrophic forgetting in connectionist networks , 1999, Trends in Cognitive Sciences.

[14]  Danijel Skocaj,et al.  Continuous Learning of Simple Visual Concepts Using Incremental Kernel Density Estimation , 2016, VISAPP.

[15]  Michael E. Hasselmo,et al.  High acetylcholine sets circuit dynamics for attention and encoding ; Low acetylcholine sets dynamics for consolidation , 2004 .

[16]  Heiko Wersing,et al.  A biologically motivated visual memory architecture for online learning of objects , 2008, Neural Networks.

[17]  L. Squire,et al.  The medial temporal lobe memory system , 1991, Science.